On Fri, Aug 02, 2002 at 03:36:20PM -0400, Eric Comeau wrote:
> Has anyone implemented the HA solution linked
> <http://www.taygeta.com/ha-postgresql.html ) linked> on the PostgreSQL
> Technical Documentation Web site?
>
> http://www.taygeta.com/ha-postgresql.html
> <http://www.taygeta.com/ha-postgresql.html>
>
> I have read through the article, but one concern I have is that I don't
> see how this is any different than performing a "File system level
> backup".
It's not different, and such a solution should not be recommended.
The procedure is broken, but there are a couple of ways to fix it:
1. Take a snapshot before running the rsync. This will ensure that
the backup-server gets a copy of files that are "in sync" rather
than a set of files copied at slightly different times. There was
a long discussion about filesystem-backup, snapshots and simular
issues on this list a few weeks ago - you might want to check it
out.
2. Use shared storage rather than rsync. (IOW: a RAID connected
to both servers at the same time).
3. Use rserv (the replication-service included in postgresql-contrib)
instead of rsync.
Both 1 and 3 will fix the consistancy-problem, but the backup-server
would only be up-to-date right after you do the sync. So, you risk
loosing several minutes of data.
Shared storage would eliminate this totally. The downside is that since
you only have one storage-unit it may become a single point of failure.
This can be fixed by using a storage-system with redundant raid-controllers,
multiple fans and multiple power-supplies. You would also want to take
regular backups, just in case.
> How is taking Rsync copies ever 5 or 10 minutes any different than
> taking a file system backup without shutting the database down?
It's not.
--
Ragnar Kjørstad
Big Storage
